Why Workplace Wellbeing AI Needs a New Ethics of Consent

Across the U.S. and globally, employers—including corporations, healthcare systems, universities, and nonprofits—are increasing investment in worker well-being. The global corporate wellness market reached $53.5 billion in sales in 2024, with North America leading adoption. Corporate wellness programs now use AI to monitor stress, track burnout risk, or recommend personalized interventions.

Vendors offering AI-enabled well-being platforms, chatbots, and stress-tracking tools are rapidly expanding. Chatbots such as Woebot and Wysa are increasingly integrated into workplace wellness programs.

Recently, Indian health platform Tata 1mg partnered with payroll fintech OneBanc to integrate AI-driven corporate healthcare directly into payroll systems, embedding wellness analytics into routine employment infrastructure rather than treating mental-health support as a separate benefit. Similar deployments are emerging across sectors.

While no public data reliably quantify how many workers use AI wellness tools, market growth and vendor proliferation suggest these systems already reach millions of workers. The market for chatbot-based mental-health apps alone is estimated at $2.1 billion in 2025, projected to grow to $7.5 billion by 2034.

Observers report that AI can potentially enhance workplace wellness by analyzing patterns of employee fatigue, scheduling micro-breaks, and flagging early signs of overload. Tools such as Virtuosis AI can analyze voice and speech patterns during meetings to detect worker stress and emotional strain.

On the surface, these technologies promise care, prevention, and support.

Imagine your supervisor asking, “Would you like to try this new AI tool that helps monitor stress and well-being? Completely optional, of course.”

The offer sounds supportive, even generous. But if you are like most employees, you do not truly feel free to decline. Consent offered in the presence of managerial power is never just consent—it is a performance, often a tacit obligation. And as AI well-being tools seep deeper into workplaces, this illusion of choice becomes even more fragile.

The risks are no longer hypothetical: Amazon has faced public criticism over wellness-framed, productivity-linked workplace monitoring, raising concerns about how well-being rhetoric can justify expanding surveillance.

At the center of this tension is the ideal of informed consent, which for decades has been the ethical backbone of data collection. If people are told what data is gathered, how it will be used, and what risks it carries, then their agreement is considered meaningful. But this model fails when applied to AI-driven well-being tools.

First, informed consent assumes a single and static moment of agreement, while AI systems operate continuously. A worker may click “yes” once, but the system collects behavioral and physiological signals throughout the day—none of which were fully foreseeable when the worker agreed. It seems unfair that consent is a one-time act, yet the data collection continues indefinitely.

Second, the information that workers receive during consent is often inadequate, vague, or too complex. Privacy notices promise that data will be “aggregated,” “anonymized,” or used to “improve engagement”—phrases that obscure the reality that AI systems generate inferences about mood, stress, or disengagement. Even when disclosures are technically correct, they are too complex for workers to meaningfully understand. Workers end up consenting amidst power inequities and socio-organizational complexities.

And then there is consent fatigue. Workers face constant prompts—policy updates, cookie banners, new app permissions. Eventually, one might click “yes” simply to continue working. Consent would rather become a reflex or convenience rather than a choice.

To be sure, workplaces have made meaningful progress in supporting well-being, and AI can genuinely help when implemented thoughtfully.

Many organizations have expanded mental health benefits and adopted flexible or hybrid work models shown to reduce stress and improve work–life balance. Likewise, empirical research suggests AI can indirectly enhance well-being by improving task optimization and workplace safety.

Such advances in workplace AI tools are critical. Yet even with expanded structural support and promising technologies, the mindset around work and worker expectations has not kept pace—shaping how well-being tools are experienced and often making workers feel compelled to say yes, even when framed as “optional.”

Even perfect consent notices cannot overcome workplace power. Workers know that managers control evaluations, promotions, and workloads. Declining a “voluntary” well-being tool can feel risky, even if the consequences are unspoken. Consent becomes a reflection of workplace politics rather than an expression of personal autonomy.

Drawing from feminist theories of sexual consent, the FRIES model of affirmative consent-- Freely given, Reversible, Informed, Enthusiastic, and Specific—provides a sharp lens for evaluating workplace use of AI.

Consent is not freely given when declining feels risky. It is not reversible when withdrawing later invites scrutiny. It is not informed when AI inference is opaque or evolving. It is rarely enthusiastic; many workers say yes out of self-protection. And it is almost never specific; opting into a single function often authorizes far more data collection than workers realize.

This is where the FRIES model offers clarity, echoing the feminist, sex-positive shift from a “no means no” standard to a “yes means yes” understanding of consent. Consent is not freely given when declining feels unsafe.

It is not reversible when opting out later raises questions. It is not informed when AI inference is opaque. It is rarely enthusiastic; many say yes to avoid negative assumptions. And it is almost never specific; agreeing to one feature often enables a broader system of hidden data tracking.

In our own research on workplace well-being technologies, workers stressed that meaningful consent requires changes not only to the technology but to the policies and organizational practices around it—underscoring that workplace consent is a structural problem—something that requires socio-technical solutions, not just better disclosure screens.

If employers want meaningful consent, they must move beyond checkbox compliance and create conditions where affirmative and continuous consent is truly possible. Participation must be genuinely voluntary.

Opting out must have no social or professional penalty—neither explicit nor implicit. Data practices need to be transparent and auditable. Most importantly, well-being must be grounded in organizational culture—not in the hope that an algorithm can fix structural problems or unrealistic expectations.

The real challenge is not perfecting AI that claims to care for workers but building workplaces where care is already embedded—where consent is real, autonomy is respected, and technology supports people.

Dr. Koustuv Saha is an Assistant Professor of Computer Science at the University of Illinois Urbana-Champaign’s (UIUC) Siebel School of Computing and Data Science and is a Public Voices Fellow of The OpEd Project.